The playbook, published August 21, divides preparation into three stages: being discovered by an agent, being selected at the point of decision, and measuring what happens after an AI-assisted referral. MediaPost first reported details from Microsoft’s marketer briefing, including the proposed 90-day activation structure. Microsoft’s own publication confirms the three-part program and the 90-day roadmap, though its public page does not spell out the reported 45-day/45-day sequencing in detail.
For businesses already running Microsoft Merchant Center feeds or Microsoft Advertising campaigns, the practical consequence is clear: product-feed management is being recast from a shopping-ad requirement into a prerequisite for inclusion in Copilot-led product discovery. For developers, the change makes commerce data quality, payment integration, and analytics instrumentation part of the same deployment conversation rather than separate marketing tasks.
Product data becomes the agent’s shortlist
Microsoft’s playbook centers on a premise that deserves scrutiny: agents consume structured information rather than judging a storefront the way a human browser does. In practice, that means title consistency, precise product attributes, current pricing, stock status, variant identifiers, fulfillment restrictions, and return terms carry more weight than a polished landing page that leaves key facts ambiguous.
Microsoft specifically points merchants toward an AI-ready Microsoft Merchant Center feed. The company says that feed helps agents identify and trust what a seller offers. That is a material shift for retailers that have treated feeds as a maintenance chore delegated to a marketing vendor. A feed with missing GTINs, mismatched landing-page prices, unavailable variants, or generic attributes can produce bad shopping-ad results today; in an agentic flow, it may keep the item from entering the recommendation set.
The company’s language also points to an uncomfortable limitation of current “AI optimization” advice. Microsoft can tell merchants to become discoverable, but it does not publish a ranking formula, a guaranteed inclusion threshold, or a public list of all the signals Copilot uses to select products. The playbook is therefore a readiness guide, not a promise of traffic or placement.
Microsoft cites its own 2025 U.S. consumer survey saying 72% of respondents expect AI-assisted shopping experiences from retailers within a year. It also cites Adobe Analytics for a claim that AI-referred visitors converted 42% better than non-AI traffic in the first quarter of 2026. Those figures may be useful directional signals, but they are not evidence that every retailer will see the same performance. Conversion rates depend heavily on category, price, returning-customer behavior, attribution windows, and whether the AI referral is counted as a new session or a continuation of an established shopping journey.
Checkout is the point where the blueprint becomes infrastructure
The second phase—being chosen—ties discovery to Copilot Checkout, ads across Bing and Copilot, and Microsoft’s Brand Agents offering. Microsoft had already announced in January that Copilot Checkout was beginning a U.S. rollout on Copilot.com with PayPal, Shopify, and Stripe partner activation, while non-Shopify merchants could apply for onboarding. It also said Merchant Center participation was not mandatory to sell through Copilot Checkout, although product feeds help inform organic Copilot shopping results.
That distinction matters for IT and ecommerce teams. A merchant may technically be able to participate in checkout without a Merchant Center feed, while still losing visibility where Copilot decides which products to surface. The commerce stack therefore needs two separate reviews: whether the business can transact in the flow, and whether its product information is sufficiently usable to be found before the transaction.
MediaPost’s report describes Microsoft’s plan as using a “Universal Commerce Protocol,” and says Target, Ulta Beauty, Stripe, Google, and Shopify are associated with that framework. Microsoft’s own January announcement uses different terminology: Agentic Commerce Protocol, or ACP. Microsoft’s August playbook page does not identify a protocol at all. Until Microsoft publishes a technical document that reconciles those names and lists the supported integrations, merchants should not assume that a claimed protocol affiliation means their existing catalog, checkout, or payment implementation is automatically compatible with Copilot.
That omission is more than semantic. Protocol compatibility determines where a retailer must expose inventory and payment data, what permissions an agent can receive, how orders are confirmed, and who owns the customer-service handoff if an item is canceled or substituted. Those are implementation details that a marketing playbook does not settle.
Clarity’s new role is measuring AI referrals
The final stage moves from traffic acquisition to observability. Microsoft says AI Visibility insights in Microsoft Clarity can show which AI platforms cite a brand, the grounding queries and topics associated with it, comparisons with competitors, and behavior after an AI-referred visitor reaches the site. The company is framing this as support for generative engine optimization, or GEO.
For site owners, this expands the purpose of Clarity from behavioral analytics—session replays, heatmaps, engagement, and conversion review—into a measurement layer for AI-mediated discovery. Microsoft previously described Brand Agents insights in Clarity as a way to compare agent-assisted sessions with organic traffic and monitor metrics including engagement, conversion uplift, and average order value.
There is a meaningful measurement caveat. Seeing that an AI platform cited a brand does not by itself prove that citation drove a sale, and a conventional analytics system cannot always reconstruct the full reasoning path inside a third-party agent. Merchants will still need clean UTM conventions where available, reliable conversion events, server-side order records, and a way to separate assisted purchases from direct, paid-search, and returning-customer revenue.
Organizations should also treat Microsoft’s visibility metrics as product telemetry, not neutral market measurement. Clarity can be useful for deciding whether Copilot-related traffic behaves differently on a site, but it is not an independent census of every AI answer or every competing shopping agent.
AI Max extends the same automation model into search campaigns
The playbook lands days after Microsoft Advertising began a global rollout of AI Max for Search campaigns. Search Engine Journal reports that AI Max expands search-term matching beyond an advertiser’s keyword list, uses site and campaign content to tailor ad text, and can route a visitor to a page other than the campaign’s default final URL when Microsoft’s systems judge it a better match.
Those capabilities mirror the playbook’s broader model: give Microsoft more structured content and conversion data, then let its systems decide which query, message, product page, or transaction route is most relevant. AI Max includes controls that matter for risk management, including brand inclusions and exclusions, term exclusions for generated text, and URL rules that constrain final-URL expansion. Existing campaigns already using Microsoft’s predictive matching or autogenerated assets may find the corresponding AI Max settings enabled, according to Search Engine Journal; the rest of the feature set requires opt-in.
Advertisers should resist the temptation to enable broad automation and call the job finished. Before testing AI Max alongside an AI-shopping effort, they need to verify that excluded products, regulated claims, discontinued SKUs, regional inventory rules, and unwanted landing pages are represented in the campaign controls. An AI-selected URL is only useful if the page accurately reflects the offer an agent or ad promised.
Microsoft is also scheduled to remove Max CPC controls from new non-portfolio campaigns using several automated bid strategies beginning October 1, 2026, according to Search Engine Journal’s reporting on the change. Existing campaigns that already use Max CPC will retain it, as will portfolio bid strategies. The timing makes the playbook’s focus on conversion data more consequential: advertisers being asked to hand more bidding discretion to the platform will need trustworthy revenue, margin, lead-quality, and refund signals to keep automated systems pointed at outcomes that actually matter.
The first 90 days should expose gaps, not merely add tags
The strongest part of Microsoft’s blueprint is its insistence that agentic shopping is a cross-functional problem. Product teams control catalog truth; developers own site performance, checkout reliability, and API integrations; marketers run feeds and campaigns; analysts determine whether referrals produced profitable orders. Any retailer assigning this solely to an SEO or paid-media team is likely to discover defects only after an agent has surfaced wrong prices or routed shoppers to unavailable products.
A useful first pass is to inventory every field that can change a purchasing decision, compare the feed against the product page and order system, test checkout edge cases, and document how an agent-assisted order will be attributed and supported. Microsoft’s playbook provides a vendor-specific route into that work. It does not remove the need to verify the plumbing.
The immediate milestone is not a promised wave of Copilot sales. It is whether merchants can make their product and transaction data consistent enough that a Microsoft-controlled agent can reliably discover, recommend, and complete a purchase without introducing a pricing, inventory, attribution, or customer-service failure.